Pushing GPT’s Creativity to Its Limits: Alternative Uses and Torrance Tests

In this paper, we investigate the potential of Large LanguageModels  (LLMs),  specifically  GPT-4,  to  improve  their  cre-ative responses in well-known creativity tests, such as Guil-ford’s Alternative Uses Test (AUT) and an adapted version ofthe Torrance Test of Creative Thinking (TTCT) visual com-pletion tests.  We exploit GPT-4’s self-improving ability byusing a sequence of forceful interactive prompts in a multi-step conversation, aiming to accelerate the convergence pro-cess towards more creative responses.  Our contributions in-clude an automated approach to enhance GPT’s responses inthe  AUT  and  TTCT  visual  completion  test  and  a  series  ofprompts to generate and evaluate GPT’s responses in thesetests. Our results show that the creativity of GPT’s responsescan be improved through the use of forceful prompts.  Thispaper opens up possibilities for future research on differentsets of prompts to further improve the creativity convergenceof LLM-generated responses and the application of similarinteractive processes to tasks involving other cognitive skills.